`session-orchestrator`

Structured multi-agent development sessions for Claude Code & Codex CLI — wave-based execution, mechanical guardrails, persistent memory, and honest self-measurement.

Wave executionDiscovery → Impl → Quality → Finalization with parallel sub-agents, scope enforcement, and inter-wave quality gates.

Guardrails, not vibesDestructive-command guard, file-scope hooks, circuit breakers, verification-before-completion discipline.

Memory that persistsSession records, learnings, and agent-proposed insights — carried across sessions, mirrored to your knowledge vault.

Honest evalaiat-llm-eval v1Every session can be scored against a pre-registered rubric — deterministic first, per-dimension, no global score, offline re-verifiable.

Get it

/plugin marketplace add Kanevry/session-orchestrator
/plugin install session-orchestrator@kanevry

Source, docs, and issues: github.com/Kanevry/session-orchestrator

The eval standard

aiat-llm-eval v1 — an open standard for honest LLM session evaluation: pre-registration, deterministic-before-judge, three-state verdicts with explicit abstention, full version provenance, and reproducibility as an executable proof (--verify), never a claim.

Leaderboard coming later — the record schema is ready; the site is deliberately not.